individual variability in functional connectivity ... · previous human fcmri studies demonstrated...
TRANSCRIPT
1
Individual variability in functional connectivity architecture of the mouse
brain
Eyal Bergmann, Xenia Gofman, Alexandra Kavushansky, Itamar Kahn*
Department of Neuroscience, Rappaport Faculty of Medicine and Institute, Technion – Israel Institute of
Technology, Haifa, Israel
*Corresponding author: Itamar Kahn, Rappaport Faculty of Medicine, Technion – Israel Institute of
Technology, 1 Efron St., POB 9649 Bat Galim, Haifa 31096, Israel; [email protected]
Contributions
E.B. and I.K. designed the study; E.B. collected the MRI data; A.K collected the behavioral data; E.B. and
X.G. analyzed the data; E.B and I.K. prepared the manuscript.
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 18, 2020. . https://doi.org/10.1101/2020.03.16.993709doi: bioRxiv preprint
2
Abstract
The functional organization of brain networks can be estimated using fMRI by examining the
coherence of spontaneous fluctuations in the fMRI signal, a method known as resting-state
functional connectivity MRI. Previous studies in humans reported that such functional networks
are dominated by stable group and individual factors, demonstrating that fMRI is suited to
measuring subject-specific characteristics, and suggesting the utility of such precision fMRI
approach in personalized medicine. However, mechanistic investigations to the sources of
individual variability in health and disease are limited in humans and thus require animal models.
Here, we used repeated-measurement resting-state fMRI in awake mice to quantify the contribution
of individual variation to the functional architecture of the mouse cortex. Comparing the
organization of functional networks across the group, we found dominant common organizational
principles. The data also revealed stable individual features, which create a unique fingerprint that
allow identification of individual mice from the group. Examining the distribution of individual
variation across the mouse cortex, we found it is homogeneously distributed in both sensory and
association networks. Finally, connectome-based predictive modeling of motor behavior in the
rotarod task revealed that individual variation in functional connectivity explained behavioral
variability. Collectively, these results show that mouse functional networks are characterized by
individual variations suggesting that individual variation characterizes the mammalian cortex in
general, and not only the primate cortex. These findings lay the foundation for future mechanistic
investigations of individual brain organization and pre-clinical studies of brain disorders in the
context of personalized medicine.
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 18, 2020. . https://doi.org/10.1101/2020.03.16.993709doi: bioRxiv preprint
3
Introduction
A fundamental question in brain research is what makes individuals different from each other. This
question can be addressed at different levels of organization, starting from genetics or
neurotransmitters, going through structural or functional measures of brain regions and networks,
and ending at behavioral phenotypes or clinical outcomes. In humans, a common approach to study
brain organization in individuals is resting-state functional connectivity fMRI (fcMRI), which
estimates functional connectivity between regions based on coherent spontaneous fluctuation in the
fMRI signal1–3.
Previous human fcMRI studies demonstrated that this measure is stable over time and can be used
to characterize individual differences4–8. These works revealed that such differences are spread
heterogeneously across the human cortex, demonstrating increased variability in association
networks, and correlation with distal connectivity and cortices that underwent expansion and
elaboration relative to non-human primates and lower mammals. Moreover, individual variations
in functional connectivity were shown to predict individual activity patterns in task conditions9,10
and behavioral performance11–13. Finally, recent studies in patients with neuropsychiatric disorders
reported that individual functional connectivity patterns can be used as a biomarker for diagnosis
and treatment optimization14–16, key features of personalized medicine.
While individual differences in functional connectivity were thoroughly characterized in humans,
including identification of sources of intra-subject variability17, a dissection of mechanisms relies
on animal models. Such investigation demands adequate sample size that is hard to achieve in
studies in non-human primates and may involve genetic manipulations and molecular techniques
that are more readily accessible in rodent models, and specifically in mice. Previous fcMRI studies
in anesthetized mice demonstrated reproducible resting-state networks18,19, applications to mouse
models of neuropsychiatric diseases20–22, and correlations between functional connectivity and
behavioral measures23–26. However, characterization of individual differences in functional
connectivity is based on repeated data acquisition that can control for measurement instability.
Since this experimental design is hard to achieve in anesthetized animals, such studies are more
suitable to awake mouse imaging. We have previously established fcMRI experiments in awake
head-fixed mice27,28 and used repeated-measurement designs to link individual differences in
structural and functional connectivity29. However, a detailed analysis of individual variability of
functional connectivity in the mouse brain and its relevance to behavior has heretofore not been
demonstrated.
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 18, 2020. . https://doi.org/10.1101/2020.03.16.993709doi: bioRxiv preprint
4
Here we used repeated-measurement resting-state fMRI to characterize individual variation in
functional connectivity in the mouse cortex. We show that despite the mouse cortex reduced
complexity relative to the human homologue and the animals being genetically identical, it is also
characterized by individual variation, allowing identification of specific mice from a group. Then,
we characterize factors affecting identification accuracy, and examine the distribution individual
variability in sensory and association networks. Finally, we link individual differences in functional
connectivity to behavioral variability in the rotarod task. Collectively, these findings indicate that
mouse functional networks are characterized by behaviorally relevant individual variation and lay
the foundation for mechanistic investigations of sources of individual variability and pre-clinical
studies of brain disorders in the context of personalized medicine.
Results
Connectivity-based identification of individual mice
Data of the study consisted of nineteen F1 C6/129P (male, age 9–12 weeks), which underwent
multiple fcMRI sessions during passive wakefulness as previously described29, followed by
behavioral testing in the rotarod task30. After exclusion of sessions with image artifacts or excessive
motion (see methods), the final dataset included 16 mice with six sessions (each comprising ~30
minutes of data), which were split to two halves of three sessions each to examine the group and
individual similarities in the mouse functional connectome. For the comparison between functional
connectivity and behavior, two additional mice with 4–5 sessions were included. In this analysis,
data from all sessions were averaged, resulting in a single connectivity matrix per mouse.
Functional connectivity matrices were built based on the Allen Common Coordinate Framework
reference atlas (CCFv3) downsampled to fMRI resolution (Fig. 1A), including 43 cortical parcels
per hemisphere, which were divided to six modules (Prefrontal, Lateral, Somatomotor, Visual,
Medial and Auditory) based on anatomical connectivity patterns31. The two connectivity matrices
of each mouse were compared to each other and to all other connectivity matrices of the other mice
(representative matrices are presented in Fig. 1B) by calculating the Fisher-transformed
correlations between all edges in the connectome. The resulting network similarity matrix (Fig.
1C) diagonal represents the similarity within individual, while rows and columns represent
similarity between specific mouse and all other mice in the group. This matrix was used for
quantification of individual variation in the connectome by comparing group and individual
network similarities7. We found substantial group similarity (mean z(r) = 0.9), which indicates
mouse connectomes share a common structure. Nevertheless, we also found that individual
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 18, 2020. . https://doi.org/10.1101/2020.03.16.993709doi: bioRxiv preprint
5
similarity (mean z(r) = 1.08) is significantly higher than group similarity (two-tailed paired student
t-test: t(15) = 7.32, P < 0.001, Cohen’s d = 1.83), demonstrating larger values in all mice (Fig. 1D),
and indicating that fcMRI can capture individual variability in the mouse connectome.
The significant difference between group and individual network similarities means that on
average, connectivity matrices from the same mouse are more similar than connectivity matrices
from different mice. A more stringent criterion for connectome-based fingerprinting is successful
identification of specific mouse from a group12. Namely, that the similarity between the two
connectivity matrices of the same mouse must be higher than the similarity to all other connectivity
matrices in the group. To test the feasibility of functional connectome-based identification in mice,
we calculated the fraction of mice in which the values along the diagonal of the similarity matrix
were higher than other values in each row and column. We found that the rate of successful
identification of mice in the first and second halves of data was 68.75% and 62.5% out of the 16
mice, respectively (Fig. 1E). To formally test the significance of those rates, we used a shuffling
procedure, calculated the distribution of identification rates when assigning random identities in
either the first or second halves of connectivity matrices, and found that the observed identification
rates are significantly higher than values in all 1000 shuffled iterations (P < 0.001, Fig. 1F). This
result indicates that that the mouse functional connectome acts as a unique fingerprint.
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 18, 2020. . https://doi.org/10.1101/2020.03.16.993709doi: bioRxiv preprint
6
Figure 1. Functional connectome fingerprinting in the mouse cortex. (A) Anatomical parcellation of the
mouse cortex overlaid on average fMRI image at native resolution. Eighty six cortical labels (black borders)
were taken from the Allen Mouse Brain Atlas CCFv3 and divided to six modules (label color) based on their
anatomical connectivity profiles31. (B) Individual functional connectomes of three representative mice from
the cohort demonstrate unique patterns that are consistent between the first and second halves of data. The
identity of the anatomical modules of each node is designated by the color. (C) A network similarity matrix
in which each cell represents the similarity between two connectomes. The matrix is asymmetric as columns
represent the similarity between the first half of data of each mouse and the second half of the data of all the
other mice, while rows represent the similarity between the second half of data of each mouse and the first
half of data of all the other mice. The values along the diagonal represent individual similarity, which is the
correlation between the edges of the two connectivity matrices of the same mouse. (D) A comparison between
group and individual similarities demonstrate higher similarity between connectomes from the same mouse
(***P < 0.001), the magenta line depicts the average group and individual network similarities. (E) True
connectome-based identification rates (magenta) were compared to 1000 iterations of identity shuffling
indicating that the mouse functional connectome acts as a unique fingerprint.
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 18, 2020. . https://doi.org/10.1101/2020.03.16.993709doi: bioRxiv preprint
7
Factors contributing to characterization of individual variation
Studies in humans demonstrated that successful identification12 and precise characterization of
individual variation7 depend on the amount of fcMRI data available per participant. Therefore,
leveraging the repeated-measurement design of our fcMRI experiment, we sought to characterize
how much data is needed to stably characterize the functional cortical organization in individual
mice.
In our original analysis (Fig. 1) two average connectivity matrices were calculated per mouse by
splitting its six sessions to two halves, controlling for the total number of included frames per half
(see Methods). In the current analysis we built a set of connectivity matrices using all combinations
of one (n = 15), two (n=45) or three sessions (n=20) per mouse, and examined network similarity
and identification rates as a function of the number of sessions averaged per connectome (Fig. 2a).
First, we calculated group and individual network similarity values for different number of included
sessions and submitted the results to a repeated-measures ANOVA (corrected with the Huynh-Feldt
method) with Individuality and Number of Sessions as within mouse factors. We found significant
effects of both factors (Individuality: F(1, 15) = 34.28 P < 0.001, H-F = 1, η2 = 0.71; Number of
Sessions: F(2, 30) = 7318.76, P < 0.001, H-F = 0.51, η2 = 0.998), confirming that network similarity
between connectivity matrices from the same mouse is higher than group similarity, and that
increasing the amount of data per mouse improves network similarity estimation. Importantly, we
also found a significant interaction between Individuality and Number of Sessions (F(2, 30) = 109.87,
P < 0.001, H-F = 0.5, η2 = 0.887), indicating that increasing the amount of data per mouse
preferentially increases individual over group network similarity values. In agreement with this
finding, the results of the identification analysis revealed that increasing the number of sessions
improve identification (two-tailed unpaired student t-test: two sessions vs. one session: t(58) = 9.08,
P < 0.001, Cohen’s d = 2.9; three sessions vs. one session: t(33) = 15.53, P < 0.001, Cohen’s d = 5.3;
three sessions vs. two sessions: t(63) = 8, P < 0.001, Cohen’s d = 2.15). Collectively, these analyses
indicate that repeat-measurement fcMRI designs supported by awake imaging are useful for
characterizing individual functional connectivity patterns.
After establishing that individual functional connectivity profiles act as a unique fingerprint in
mice, we sought to characterize which brain connections contribute to individual variation.
Therefore, we replicated that analysis of Finn et al.12 and characterized which edges possess high
differential power (DP) that contribute to identification, and which edges possess high group
consistency (Φ) within a mouse and across the group. We derived those values for all edges in the
connectivity matrix, and determined which edges were in the top 1% of each measure (Fig. 2b).
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 18, 2020. . https://doi.org/10.1101/2020.03.16.993709doi: bioRxiv preprint
8
We found that most edges in the top 1% of DP are related to the posterior modules, namely, Visual,
Medial and Auditory, and include many inter-hemispheric and inter-module connections. In
contrast, edges in the top 1% of Φ came from all modules and included mainly intra-hemispheric
intra-module connections. In the original analysis of Finn et al.12, high DP were observed in high-
order fronto-parietal connections, while high Φ were observed in inter-hemispheric connectivity
within the Somatomotor and Visual networks. Comparing the analyses across species, we conclude
that while the mouse data recapitulate the Φ difference between intra- and inter-module
connections, and show DP bias towards specific modules, it seems to not show a difference between
sensory and association regions. Therefore, we sought to directly examine individual variation in
those qualitatively different brain systems.
Figure 2. Factors contributing to characterization of individual variation. (A) Network similarity values
(top) and identification rates (bottom) as a function of number of included sessions in the two connectivity
matrices of each mouse. All combinations of data sampling of one, two or three sessions were calculated and
compared using repeated-measures ANOVA (network similarity, see text) or two-tailed unpaired student t-
test (identification rate, *** P < 0.001). error bars (top) represent the standard error of the mean; boxplots
(bottom) represent the median (center line), interquartile range (box limits); 1.5 × interquartile range
(whiskers) and outlier (points). (B) Unique (DP, top) and consistent (Φ, bottom) edges in individual
connctomes. For visualization purposes, both measures were thresholded at the 99th percentile. In the circle
plot (left) the 86 nodes are organized based on their anatomical module identity and cortical hemisphere;
lines indicate edges. In the matrices (right), the fraction of edges connecting between and within modules is
color coded with darkly shaded cells representing higher DP (top) and Φ (bottom) values. PF, prefrontal; Lat,
lateral; SM, somatomotor; VIS, visual; MED, medial; AUD, auditory.
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 18, 2020. . https://doi.org/10.1101/2020.03.16.993709doi: bioRxiv preprint
9
Individual variation across brain systems
To better characterize individual variation in sensory and association systems in the mouse cortex,
we sought to examine whether functional connectivity profiles within these systems differ in their
identifiability or group and individual network similarities. Therefore, we defined each cortical
node as either sensory (Somatomotor, Visual, and Auditory modules) or association (Prefrontal,
Lateral, and Medial modules) and derived two network similarity matrices (Fig. 3A). Then, we
calculated group and individual network similarities for each system (Fig. 3B), and found higher
values compare to the original full connectivity matrix (all connections) in both group (two-tailed
paired student t-test: all connections vs. association: t(15) = 26.9, P < 0.001, Cohen’s d = 6.73; all
connections vs. sensory: t(15) = 18.44, P < 0.001, Cohen’s d = 4.61) and individual similarities (all
connections vs. association: t(15) = 8.92, P < 0.001, Cohen’s d = 2.23; all connections vs. sensory:
t(15) = 11.59, P < 0.001, Cohen’s d = 2.9). Comparison between association and sensory systems
reveal significant difference in group network similarity (t(15) = 4.96, P < 0.001, Cohen’s d = 1.24),
but no difference in individual network similarity (t(15) = 0.56 P, = 0.583, Cohen’s d = 0.14; all
values were corrected for multiple comparison using false-discovery rate based on the Benjamini-
Hochberg method). Examining the normalized relative effect magnitude of individuality (Fig. 3C),
we found no difference between sensory and association systems (t(15) = 1.803, P = 0.137, Cohen’s
d = 0.45). Nevertheless, comparison to the full connectivity matrix revealed lower effect magnitude
of individuality in association (t(15) = 3.02, P = 0.026, Cohen’s d = 0.75), but not sensory (t(15) =
0.134, P = 0.85, Cohen’s d = 0.03), systems. Finally, we carried out the identification analysis for
each type of connectome and found similar identification rates in sensory and association systems
(Fig. 3D), which were only slightly lower than the original values yielded by the full connectivity
matrix. Collectively, these analyses indicate that connections between sensory and association
regions, which reflect inter-module connectivity, are less consistent across the group. Importantly,
the results indicate that the relative effect magnitude of individuality on functional organization of
the mouse cortex is more modest compared to humans7, and the qualitative differences between
sensory and association networks in the human brain are not well recapitulated in mice.
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 18, 2020. . https://doi.org/10.1101/2020.03.16.993709doi: bioRxiv preprint
10
Figure 3. Comparison of individual variation between association and sensory cortical systems. (A)
Segmentation of cortical modules to association and sensory networks (left) and the similarity matrices
between individual connectomes which were limited to either association or sensory regions (right). (B)
Comparison between group and individual similarities of the full connectome (All) and the connectomes
built using only association (Assoc) or sensory regions demonstrates that the full connectome is characterized
by lower similarity values at both Group and Individual levels (***P < 0.001). Boxplots represent the median
(center line), interquartile range (box limits) and 1.5 × interquartile range (whiskers). (C) Comparison
between normalized relative individual effect magnitude of the different connectomes reveals no difference
between sensory and association networks. Comparison to the full connectome revealed that it is
characterized by a stronger effect of individuality relative to association networks (*P < 0.05). (D)
Identification rates of individual mice using the three different connectomes, demonstrate slightly increased
performance in the full connectome with minimal differences between association and sensory networks. h1,
first half of data; h2, second half.
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 18, 2020. . https://doi.org/10.1101/2020.03.16.993709doi: bioRxiv preprint
11
Individual brain-behavior relations in mice
After characterizing individual variation in functional connectivity in the mouse cortex, we sought
to examine whether it can predict behavioral phenotypes. Therefore, we used connectome-based
predictive modelling (CPM)32 to link between functional connectivity profiles and behavioral
performance in the rotarod task.
Examining rotarod performance, we found prominent variability within the group with different
mice presenting wide range of latencies to fall (Fig. 4A). This behavioral variability could be
predicted by the functional connectivity data using CPM (Fig. 4B) as leave-one-out cross-
validation (LOOCV) analysis demonstrated good correspondence (r(16) = 0.51) between predicted
and observed mean latencies to fall. To formally test the goodness of prediction, we shuffled the
behavioral data 1000 times and ran LOOCV analyses on shuffled data to extract significance level
(P = 0.021), which confirmed significant prediction. Importantly, latency to fall values were not
correlated with individual average head motion estimates during scanning (r(16) = -0.27, P = 0.27),
confirming that the prediction is not artifactually increased by motion patterns.
Finally, we explored which functional connections contributed to the model (Fig. 4C) and found
that positive correlations were more frequent in edges connecting sensory nodes, while negative
correlations were less frequent and characterized mainly edges connecting between sensory and
lateral association nodes. To formally test this observation we used a set of two-tailed Z-tests for
independent proportions (Fig. 4D) to test whether significant edges are biased towards connections
between association nodes (A:A, n = 946), sensory nodes (S:S, n = 861) or one sensory and one
association node (A:S, n = 1848, all comparisons were corrected for multiple comparisons using
false-discovery rate). This analysis confirm that positive correlations are more frequent in S:S
connections than in A:S (Z = 5, P < 0.001) and A-A (Z = 6.07, P < 0.001) connections, with
additional bias towards A:S relative to A:A connections (Z = 2.62, P = 0.009). In contrary, negative
correlations were more frequent in A:S connections, demonstrating significant bias relative to S:S
connections (Z = 2.27, P = 0.023), and marginally significant bias relative A:A connections (Z =
1.87, P = 0.061); comparison of A:A and S:S connections revealed no difference (Z = 0.95, P =
0.34). Collectively, the data establish that fcMRI can be used to characterize individual brain-
behavior relations.
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 18, 2020. . https://doi.org/10.1101/2020.03.16.993709doi: bioRxiv preprint
12
Figure 4. Connectome-based predictive modeling of individual performance in the rotarod task. (A)
individual (light gray) and group average (magenta) performance in the rotarod task (top), demonstrate
considerable variability in the mean latency to fall of different mice in the group (bottom). (B) Relations
between observed and predicted mean latency to fall values show close agreement (top). A formal statistical
analysis was done using a shuffling analysis in which connectome-based predictive modeling was applied to
shuffled rotatrod data, the original r-value (magenta arrow) designates the 21 highest value out of 1000
iterations. (C) Circle plot (left) and matrix (right) representations of edges that were positively (red, top) or
negatively (blue, bottom) correlated with rotarod performance in all leave-one-out cross validation iterations.
(D) All edges from the previous analysis were divided to three categories, connections between two
association regions (assoc:assoc), connections between an association and a sensory region (assoc:sensory)
and connections between two sensory regions (sensory:sensory). The results revealed that edges which are
positively correlated with the task are biased to sensory:sensory connections (red, top), while edges that are
negatively correlated with the task are slightly biased to association:sensory connections (*P < 0.05, **P <
0.01, ***P < 0.001). PF, prefrontal; Lat, lateral; SM, somatomotor; VIS, visual; MED, medial; AUD,
auditory.
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 18, 2020. . https://doi.org/10.1101/2020.03.16.993709doi: bioRxiv preprint
13
Discussion
In this study, we characterized individual variation in the functional organization of the mouse
cortex. We found evidence for a unique and reliable functional fingerprint, that allows identification
of individual mice from a group. Then, we demonstrated that identification rates increase with the
amount of data per mouse, indicating the repeated-measurement experimental design is required
for precise characterization of mouse-specific connectomes. Comparing individual variation
between sensory and association networks, we found that the differences observed between those
cortical systems in humans are not well recapitulated in mice. Finally, we show that variance in
functional connectivity, especially between sensory cortices, can explain behavioral variability in
the rotarod task. Collectively, these findings lay the foundations for studying the functional
organization of the mouse brain in health and disease at the level of the individual animal.
While the field of precision fMRI has recently emerged in human brain research4,7,12,33,34, rodent
data are still analyzed at the group level35,36. A major challenge in precision fMRI is the need of
large amount of data per subject ranging from 50 to 100 minutes based on the studied brain
structure37. Such amount of data has yet to be achieved in anesthetized mice in which acquisition
is predominantly between several minutes and up to 40 minutes per session38,39 and a substantive
repeated-measurement design is not feasible due to the difficulties that arise from repeated
ventilation and catheterization and the potential impact of prolonged anesthesia. On the other hand,
experimental setup for awake fcMRI27,40–42 can support such repeated-measurement designs.
Therefore, despite the controversy on optimal scanning procedure in rodents and the lack of
standardization in the field43, awake but not anesthetized fcMRI is necessary for repeated-
measurement experimental designs that enable precision fcMRI analysis in individual mice.
Comparison between the findings of this work and the two seminal human studies of Finn12 and
Gratton7 shows that while the central findings of those studies are recapitulated in mice, there are
also some important differences. First, identification rates in our mouse cohort are lower and the
effect magnitude of individuality is more modest in mice, in which cortical organization is strongly
dominated by group shared features. Moreover, the heterogeneity in individual variation among
brain networks was not recapitulated in mice, demonstrating similar patterns in sensory and
association systems. While these differences might be explained by the complex organization of
the human cortex44, expansion of association networks in humans45,46 or qualitative differences in
anatomically homologues high-order structures27, we cannot exclude the genetic homogeneity in
our cohort reduces the contribution of induvial features, as genetics was previously shown the shape
functional connectivity47. In addition, some differences are perhaps related to the areal parcellation
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 18, 2020. . https://doi.org/10.1101/2020.03.16.993709doi: bioRxiv preprint
14
that was used for defining nodes in the connectome. In humans, connectomes are defined based on
functional areal parcellation resulted in several hundred seed regions48–51. Since such algorithms
have yet to be adapted to mice, and are beyond the scope of the current work, we used anatomical
parcellation based on the Allen Mouse Brain Atlas, which is comprised of 86 cortical labels31,52.
Previous studies in mice showed closed agreement between structural and functional connectivity
in the mouse brain27,53 including gross cortical organization54. Moreover, using the Allen Mouse
Brain Atlas we have recently shown that structural connectomes can predict their functional
counterpart29, suggesting that this anatomical parcellation is functionally relevant.
A major limitation of precision fMRI in humans is the restriction to one level of analysis, as it is
limited to large-scale organization55. As a result, it is hard to link individual variation in functional
connectomes to brain function and dysfunction. In contrast, tools available in rodents support causal
manipulation of brain activity using molecular techniques such as chemogenetics or optogenetics,
which can be combined with fMRI39,56–60. However, heretofore such analyses were restricted to the
group level. Future studies can use the individual functional connectome to predict the effect of
causal control similar to the way human resting-state fMRI is used to predict individual task fMRI
activation pattern9, linking the different levels of organization and uncovering sources of individual
variation. In addition, rodent models of disease are commonly studied using fcMRI21,22,61, which
allows direct translation to humans20,28, as well as studying the relations between functional
connectivity and behavior23–26. Such studies can utilize the approach presented here to follow the
trajectory of individual animals during development, aging or after treatment, as well as the CPM,
which provide a data-driven alternative to predict behavior based on fcMRI data32, instead of
following a hypothesis-driven behavioral correlations of functional connections between specific
seed regions.
Together, our results establish the feasibility of precision fMRI approach in studying the mouse
functional connectome, indicating that individual variation in the organization of cortical networks
is likely to extend across the mammalian class in general, and is not restricted to primates. Given
this foundation, future mouse fMRI studies can follow the human neuroimaging community by
moving from group-level inferences to the level of the individual animal. Such transition can be
highly beneficial for mechanistic investigation of brain organization, as well as for pre-clinical
studies of neuropsychiatric disorders in the context of personalized medicine.
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 18, 2020. . https://doi.org/10.1101/2020.03.16.993709doi: bioRxiv preprint
15
Data availability
Behavioral and Imaging raw data, as well as custom-written codes used in this study are available
in BIDS format on OpenNeuro, https://openneuro.org/datasets/ds002307. Codes for identification
and connectome-based predictive modeling adapted from the work of Finn et al.12 and Shen et al.32
which were previously published: https://www.nitrc.org/projects/bioimagesuite.
Acknowledgments
This research was supported by Israel Science Foundation (770/17), the Ministry of Science &
Technology, Israel & Ministry of Europe and Foreign Affairs (MEAE) and the Ministry of Higher
Education, Research and Innovation (MESRI) of France, the Adelis Foundation, and the Prince
Center. We thank Technion's Biological Core Facilities and Edith Suss-Toby for her assistance
with MRI, and the Technion Preclinical Research Authority, and Nadav Cohen for assistance with
animal care.
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 18, 2020. . https://doi.org/10.1101/2020.03.16.993709doi: bioRxiv preprint
16
References
1. Fox, M. D. & Raichle, M. E. Spontaneous fluctuations in brain activity observed with
functional magnetic resonance imaging. Nat. Rev. Neurosci. 8, 700–11 (2007).
2. Buckner, R. L., Krienen, F. M. & Yeo, B. T. Opportunities and limitations of intrinsic
functional connectivity MRI. Nat Neurosci 16, 832–837 (2013).
3. Power, J. D., Schlaggar, B. L. & Petersen, S. E. Studying Brain Organization via
Spontaneous fMRI Signal. Neuron 84, 681–696 (2014).
4. Gordon, E. M. et al. Precision Functional Mapping of Individual Human Brains. Neuron
95, 791-807.e7 (2017).
5. Kong, R. et al. Spatial Topography of Individual-Specific Cortical Networks Predicts
Human Cognition, Personality, and Emotion. Cereb. Cortex 29, 2533–2551 (2019).
6. Mueller, S. et al. Individual variability in functional connectivity architecture of the
human brain. Neuron 77, 586–95 (2013).
7. Gratton, C. et al. Functional Brain Networks Are Dominated by Stable Group and
Individual Factors, Not Cognitive or Daily Variation. Neuron 98, 439-452.e5 (2018).
8. Laumann, T. O. et al. Functional System and Areal Organization of a Highly Sampled
Individual Human Brain. Neuron 87, 657–70 (2015).
9. Tavor, I. et al. Task-free MRI predicts individual differences in brain activity during task
performance. Science 352, 216–20 (2016).
10. Cole, M. W., Ito, T., Bassett, D. S. & Schultz, D. H. Activity flow over resting-state
networks shapes cognitive task activations. Nat. Neurosci. 19, 1718–1726 (2016).
11. Smith, S. M. et al. A positive-negative mode of population covariation links brain
connectivity, demographics and behavior. Nat. Neurosci. 18, 1565–1567 (2015).
12. Finn, E. S. et al. Functional connectome fingerprinting: identifying individuals using
patterns of brain connectivity. Nat. Neurosci. 18, 1664–1671 (2015).
13. Rosenberg, M. D. et al. A neuromarker of sustained attention from whole-brain functional
connectivity. Nat. Neurosci. 19, 165–171 (2015).
14. Drysdale, A. T. et al. Resting-state connectivity biomarkers define neurophysiological
subtypes of depression. Nat. Med. 23, 28–38 (2017).
15. Boerwinkle, V. L. et al. Network‐targeted approach and postoperative resting‐state
functional magnetic resonance imaging are associated with seizure outcome. Ann. Neurol.
86, 344–356 (2019).
16. Abraham, A. et al. Deriving reproducible biomarkers from multi-site resting-state data:
An Autism-based example. Neuroimage 147, 736–745 (2017).
17. Poldrack, R. A. et al. Long-term neural and physiological phenotyping of a single human.
Nat. Commun. 6, (2015).
18. Gozzi, A. & Schwarz, A. J. Large-scale functional connectivity networks in the rodent
brain. Neuroimage 127, 496–509 (2016).
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 18, 2020. . https://doi.org/10.1101/2020.03.16.993709doi: bioRxiv preprint
17
19. Grandjean, J. et al. Common functional networks in the mouse brain revealed by multi-
centre resting-state fMRI analysis. bioRxiv 541060 (2019). doi:10.1101/541060
20. Bertero, A. et al. Autism-associated 16p11.2 microdeletion impairs prefrontal functional
connectivity in mouse and human. Brain (2018). doi:10.1093/brain/awy111
21. Grandjean, J. et al. Complex interplay between brain function and structure during
cerebral amyloidosis in APP transgenic mouse strains revealed by multi-parametric MRI
comparison. Neuroimage 134, 1–11 (2016).
22. Zerbi, V. et al. Dysfunctional Autism Risk Genes Cause Circuit-Specific Connectivity
Deficits With Distinct Developmental Trajectories. Cereb. Cortex 28, 2495–2506 (2018).
23. Pagani, M. et al. Deletion of autism risk gene Shank3 disrupts prefrontal connectivity. J.
Neurosci. 2529–18 (2019). doi:10.1523/JNEUROSCI.2529-18.2019
24. Michetti, C. et al. The Knockout of Synapsin II in Mice Impairs Social Behavior and
Functional Connectivity Generating an ASD-like Phenotype. Cereb. Cortex 27, 5014–
5023 (2017).
25. Liska, A. et al. Homozygous Loss of Autism-Risk Gene CNTNAP2 Results in Reduced
Local and Long-Range Prefrontal Functional Connectivity. Cereb. Cortex 28, 1141–1153
(2018).
26. Ash, J. A. et al. Functional connectivity with the retrosplenial cortex predicts cognitive
aging in rats. Proc. Natl. Acad. Sci. U. S. A. 113, 12286–12291 (2016).
27. Bergmann, E., Zur, G., Bershadsky, G. & Kahn, I. The Organization of Mouse and Human
Cortico-Hippocampal Networks Estimated by Intrinsic Functional Connectivity. 26,
4497–4512 (2016).
28. Shofty, B. et al. Autism-associated Nf1 deficiency disrupts corticocortical and
corticostriatal functional connectivity in human and mouse. Neurobiol. Dis. 130, (2019).
29. Melozzi, F. et al. Individual structural features constrain the mouse functional
connectome. Proc. Natl. Acad. Sci. U. S. A. 116, 26961–26969 (2019).
30. Deacon, R. M. J. Measuring motor coordination in mice. J. Vis. Exp. (2013).
doi:10.3791/2609
31. Harris, J. A. et al. Hierarchical organization of cortical and thalamic connectivity. Nature
575, 195–202 (2019).
32. Shen, X. et al. Using connectome-based predictive modeling to predict individual
behavior from brain connectivity. Nat. Protoc. 12, 506–518 (2017).
33. Sylvester, C. M. et al. Individual-specific functional connectivity of the amygdala: A
substrate for precision psychiatry. Proc. Natl. Acad. Sci. U. S. A. (2020).
doi:10.1073/pnas.1910842117
34. Greene, D. J. et al. Integrative and Network-Specific Connectivity of the Basal Ganglia
and Thalamus Defined in Individuals. Neuron (2019). doi:10.1016/j.neuron.2019.11.012
35. Grandjean, J. et al. Common functional networks in the mouse brain revealed by multi-
centre resting-state fMRI analysis. Neuroimage 205, (2020).
36. Zerbi, V., Grandjean, J., Rudin, M. & Wenderoth, N. Mapping the mouse brain with rs-
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 18, 2020. . https://doi.org/10.1101/2020.03.16.993709doi: bioRxiv preprint
18
fMRI: An optimized pipeline for functional network identification. Neuroimage 123, 11–
21 (2015).
37. Gratton, C. et al. Defining Individual-Specific Functional Neuroanatomy for Precision
Psychiatry. Biological Psychiatry (2020). doi:10.1016/j.biopsych.2019.10.026
38. Gutierrez-Barragan, D., Basson, M. A., Panzeri, S. & Gozzi, A. Infraslow State
Fluctuations Govern Spontaneous fMRI Network Dynamics. Curr. Biol. 29, 2295-2306.e5
(2019).
39. Zerbi, V. et al. Rapid Reconfiguration of the Functional Connectome after Chemogenetic
Locus Coeruleus Activation. Neuron (2019). doi:10.1016/j.neuron.2019.05.034
40. Yoshida, K. et al. Physiological effects of a habituation procedure for functional MRI in
awake mice using a cryogenic radiofrequency probe. J. Neurosci. Methods 274, 38–48
(2016).
41. Liang, Z., King, J. & Zhang, N. Anticorrelated resting-state functional connectivity in
awake rat brain. Neuroimage 59, 1190–9 (2012).
42. Tsurugizawa, T. et al. Awake functional MRI detects neural circuit dysfunction in a
mouse model of autism. Sci. Adv. 6, eaav4520 (2020).
43. Mandino, F. et al. Animal Functional Magnetic Resonance Imaging: Trends and Path
Toward Standardization. Front. Neuroinform. 13, 78 (2020).
44. Glasser, M. F. et al. A multi-modal parcellation of human cerebral cortex. Nature 536,
171–178 (2016).
45. Buckner, R. L. & Krienen, F. M. The evolution of distributed association networks in the
human brain. Trends Cogn. Sci. 17, 648–65 (2013).
46. Krubitzer, L. In search of a unifying theory of complex brain evolution. Ann. N. Y. Acad.
Sci. 1156, 44–67 (2009).
47. Glahn, D. C. et al. Genetic control over the resting brain. Proc. Natl. Acad. Sci. U. S. A.
107, 1223–1228 (2010).
48. Power, J. D. et al. Functional network organization of the human brain. Neuron 72, 665–
78 (2011).
49. Shen, X., Tokoglu, F., Papademetris, X. & Constable, R. T. Groupwise whole-brain
parcellation from resting-state fMRI data for network node identification. Neuroimage 82,
403–415 (2013).
50. Craddock, R. C., James, G. A., Holtzheimer, P. E., Hu, X. P. & Mayberg, H. S. A whole
brain fMRI atlas generated via spatially constrained spectral clustering. Hum. Brain Mapp.
33, 1914–1928 (2012).
51. Gordon, E. M. et al. Generation and Evaluation of a Cortical Area Parcellation from
Resting-State Correlations. Cereb. Cortex 26, 288–303 (2016).
52. Oh, S. W. et al. A mesoscale connectome of the mouse brain. Nature 508, 207–214
(2014).
53. Stafford, J. M. et al. Large-scale topology and the default mode network in the mouse
connectome. Proc Natl Acad Sci U S A 111, 18745–18750 (2014).
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 18, 2020. . https://doi.org/10.1101/2020.03.16.993709doi: bioRxiv preprint
19
54. Grandjean, J., Zerbi, V., Balsters, J. H., Wenderoth, N. & Rudin, M. Structural Basis of
Large-Scale Functional Connectivity in the Mouse. J. Neurosci. 37, 8092–8101 (2017).
55. D’Esposito, M. Are individual differences in human brain organization measured with
functional MRI meaningful? Proceedings of the National Academy of Sciences of the
United States of America 116, 22432–22434 (2019).
56. Grandjean, J. et al. A brain-wide functional map of the serotonergic responses to acute
stress and fluoxetine. doi:10.1038/s41467-018-08256-w
57. Desai, M. et al. Mapping brain networks in awake mice using combined optical neural
control and fMRI. J. Neurophysiol. 105, 1393–1405 (2011).
58. Lee, J. H. et al. Global and local fMRI signals driven by neurons defined optogenetically
by type and wiring. Nature 465, 788–792 (2010).
59. Weitz, A. J. et al. Thalamic Input to Orbitofrontal Cortex Drives Brain-wide, Frequency-
Dependent Inhibition Mediated by GABA and Zona Incerta Article Thalamic Input to
Orbitofrontal Cortex Drives Brain-wide, Frequency-Dependent Inhibition Mediated by
GABA and Zona Incerta. Neuron 104, (2019).
60. Bernal-Casas, D., Lee, H. J., Weitz, A. J. & Lee, J. H. Studying Brain Circuit Function
with Dynamic Causal Modeling for Optogenetic fMRI. Neuron 93, 522-532.e5 (2017).
61. Sforazzini, F. et al. Altered functional connectivity networks in acallosal and socially
impaired BTBR mice. Brain Struct. Funct. (2014). doi:10.1007/s00429-014-0948-9
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 18, 2020. . https://doi.org/10.1101/2020.03.16.993709doi: bioRxiv preprint
20
Online methods
Mice, surgical procedures, behavioral training
All procedures were conducted in accordance with the ethical guidelines of the National Institutes
of Health and were approved by the institutional animal care and use committee (IACUC) at
Technion. A detailed description of the experimental design was previously published1. Briefly, 19
first generation B6129PF/J1 hybrid mice (males, 9–12 weeks old) were implanted with MRI-
compatible head-posts and housed in reversed 12 h light/dark cycle. Then, acclimatized to awake
fMRI during passive wakefulness2 and underwent seven 45 min long awake imaging sessions,
followed by one structural MRI session under anesthesia (not used in the current study). A week
after functional MRI data were acquired, mice underwent a 2-day rotarod testing.
Image acquisition
MRI scans were performed at 9.4 Tesla MRI (Bruker BioSpin GmbH, Ettlingen, Germany) using
a quadrature 86 mm transmit-only coil and a 20 mm loop receive-only coil (Bruker). Each awake
fMRI session started with a brief anesthesia (5% isoflurane) to allow proper mounting to the
custom-made cradle2. Mice typically were alert within less than a minute. Mice had ~15 min to
fully recover from the anesthesia during scanner calibrations and acquisition of a short low-
resolution rapid acquisition process with a relaxation enhancement (RARE) T1-weighted structural
image (TR = 1500 ms, TE = 8.5 ms, RARE-factor = 4, FA = 180°, 30 coronal slices, 150 × 150 ×
450 µm3 voxels, no interslice gap, FOV 19.2 × 19.2 mm2, matrix size of 128 × 128). Then, four
spin echo EPI (SE-EPI) runs were acquired (TR = 2500 ms, TE = 18.398 ms, 200 time points, FA
= 90°, 30 coronal slices, 150 × 150 × 450 µm3 voxels, no interslice gap, FOV 14.4 × 9.6 mm2,
matrix size of 96 × 64) before mice were returned to their home cages.
Rotarod
To assess general motor function, we used the rotarod task3 in which mice (up to five at once) walk
on a rotating rod (ENV-575MA, Med Associates, St. Albans, VT) while the speed of rotation is
accelerating from 16 to 40 rounds per minute during the period of six minutes. Each mouse was
trained for two consecutive days over four trials per day with an inter trial interval of 15 min. The
latency between the beginning of each trial and falling time was calculated to extract individual
learning curves and later averaged across trials to extract a single value for overall task performance
per mouse.
MRI data preprocessing
Functional data were preprocessed as previously described1,2,4 including removal of the first two
frames for T1-equilibration effects, compensation for slice-dependent time shifts, rigid body
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 18, 2020. . https://doi.org/10.1101/2020.03.16.993709doi: bioRxiv preprint
21
motion correction, registration to a downsampled version of the Allen Mouse Brain Atlas (AMBC
CCFv3)5, and intensity normalization. At this stage, one mouse was excluded due to susceptibility
artifacts in the parietal cortex, and additional three sessions were also excluded due to ghosting
artifacts in the EPI.
After this general fMRI preprocessing, an fcMRI-specific preprocessing was performed. First, data
underwent motion scrubbing to remove motion-related artifacts6. Censoring criteria were frame
displacement of 50 µm and temporal derivative root mean square variance over voxels of 150%
inter-quartile range above the 75th percentile, with an augmented mask of one additional frame after
each detected movement and censoring of sequences with less than five included frames. Runs with
less than 50 frames and sessions with less than 192 frames (8 min) were excluded (a total of six
sessions). The average number of included sessions per mouse was 6.33 ± 0.84 (mean ± SD) and
the average total included time per session was 19.31 ± 3.67 min per session. After motion
scrubbing, preprocessing continued with demeaning and detrending, nuisance regression of six
motion parameters, ventricular and white matter signals, and their first derivatives, temporal filter
(0.009 < f < 0.08 Hz), and spatial smoothing (Gaussian kernel with FWHM of 450 µm) .
Construction of the functional connectome
To construct the functional connectome of each mouse, we used the AMBC Atlas to define 86
regions in the mouse cortex (43 per hemisphere), which were classified into six different modules
(Prefrontal, Lateral, Somatomotor, Visual, Medial and Auditory) based on their anatomical
connectivity patterns5. Labels were registered to the native fMRI resolution using the nearest
neighbor interpolation7,8. The very deep and superficial aspects of each label were removed to
minimize partial-volume effects. In addition, posterior parts of the retrosplenial and primary visual
cortices were also removed due to inconsistent registration in these areas.
After defining connectome nodes, we extracted their time courses in each session of each mouse,
and calculated the Fisher’s Z transformed Pearson correlation (r) values9, resulting in a 86 × 86
connectivity matrix per session. Then, these matrices were split to two halves of three sessions per
mouse for similarity and identification analyses (minimizing the difference in number of included
frames per half) or averaged across all sessions of each mouse for the connectome-based predictive
modelling (CPM) analysis. Mice with less than six valid sessions (n = 2) were included only in the
CPM analysis. In mice with seven valid sessions (n = 9), the session with the highest movement
was excluded from the similarity and identification analyses in order to match the amount of data
per half.
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 18, 2020. . https://doi.org/10.1101/2020.03.16.993709doi: bioRxiv preprint
22
Similarity and identification analyses
To quantify individual variation and perform connectome-based fingerprinting we adapted
algorithms developed in humans to estimate network similarity10 and identification rates11. Both
procedures are based on the construction of a similarity matrix, in which each cell is the Fisher’s Z
transformed correlation between val7ues in all 3655 edges in two connectomes, columns represent
the first half of data and rows represent the second half of data. For quantification of network
similarity, values along the diagonal represent individual similarity, which is the correlation
between the connectomes built for the same mouse from the halves of data. Group similarity is
defined as the average of values in a combined row and column vector (excluding the value along
the diagonal). Group and individual similarity values were compared using paired student t-test or
repeated-measures ANOVA after normality was tested using the Lilliefors test.
The identification procedure is a more stringent analysis for fingerprinting and quantifies the
fraction of mice in which the individual similarity is higher than any other value in their row or
column. To formally test whether the observed identification rates are significant, we compared
them to the distribution of rates in 1000 iterations in a shuffling analysis, in which we assigned
random identity to connectomes of one half of the data, and examined whether this random identity
can be detected using data from the other half of data. To test whether individual variation differs
between association and sensory networks in the mouse cortex, we constructed connectomes
limited to either association (n = 44) or sensory (n = 42) modules. Then, we ran the similarity and
identification analyses as described.
Quantifying the amount of data needed to study individual variation in mice
To characterize the amount of data needed for studying individual variation in mice, we examined
the effects of the number of sessions included in the construction of each connectome on network
similarity and identification rate. We built two connectomes per mouse using one, two or three
sessions per connectome, and examined all possible combinations. Then, we replicated the
similarity and identification analyses as described.
Quantifying edgewise contributions to identification
Edge-based analyses examine which connections contribute more to successful identification. A
detailed description of the calculation of group consistency (Φ) and differential power (DP) was
previously published11. Briefly, each connectome underwent z-normalization, and the edgewise
product was calculated for each edge in all pairs of connectomes. Group consistency values are the
average edgewise products from the comparisons of the two connectomes of each mouse. In
contrast, DP represents the empirical probability that edgewise product of two connectomes from
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 18, 2020. . https://doi.org/10.1101/2020.03.16.993709doi: bioRxiv preprint
23
the same mouse is higher than the edgewise product of two connectomes from different mice. The
top 1% of DP and Φ were presented in either circle plots or matrix plots grouped to modules 12.
Connectome-based predictive modelling (CPM)
To link individual variation in the functional connectome to behavioral variability in the rotarod
task, we followed a previously published detailed protocol12. We used a leave-one-out cross-
validation approach in which Spearman correlation was calculated between functional connectivity
in each edge in the connectome and rotarod mean latency to fall values. Then, significant (P < 0.05)
positive and negative correlations were selected, summarized to two values per mouse, which were
combined in a single regression model. This model was used for prediction of rotarod performance
in the one testing mouse in each one of the 18 iterations. Finally, the correlation between predicted
and observed behavioral measures was calculated and formally tested by comparing it to the
distribution of correlation in 1000 iterations in which rotarod performance was randomly assigned
to mice12. To characterize the edges that contributed to the prediction, we extracted the edges that
were included in the model in all iterations and presented them in circle and matrix plots. A set of
Z-tests for independent proportions was used to examine whether the contributing edges were
biased to connections between association or sensory regions.
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 18, 2020. . https://doi.org/10.1101/2020.03.16.993709doi: bioRxiv preprint
24
References
1. Melozzi, F. et al. Individual structural features constrain the mouse functional
connectome. Proc. Natl. Acad. Sci. U. S. A. 116, 26961–26969 (2019).
2. Bergmann, E., Zur, G., Bershadsky, G. & Kahn, I. The Organization of Mouse and Human
Cortico-Hippocampal Networks Estimated by Intrinsic Functional Connectivity. 26,
4497–4512 (2016).
3. Deacon, R. M. J. Measuring motor coordination in mice. J. Vis. Exp. (2013).
doi:10.3791/2609
4. Shofty, B. et al. Autism-associated Nf1 deficiency disrupts corticocortical and
corticostriatal functional connectivity in human and mouse. Neurobiol. Dis. 130, (2019).
5. Harris, J. A. et al. Hierarchical organization of cortical and thalamic connectivity. Nature
575, 195–202 (2019).
6. Power, J. D. et al. Methods to detect, characterize, and remove motion artifact in resting
state fMRI. Neuroimage 84, 320–41 (2014).
7. Jenkinson, M. & Smith, S. A global optimisation method for robust affine registration of
brain images. Med. Image Anal. 5, 143–156 (2001).
8. Jenkinson, M., Bannister, P., Brady, M. & Smith, S. Improved optimization for the robust
and accurate linear registration and motion correction of brain images. Neuroimage 17,
825–841 (2002).
9. Zar, J. H. Biostatistical Analysis. (Prentice Hall, 1996).
10. Gratton, C. et al. Functional Brain Networks Are Dominated by Stable Group and
Individual Factors, Not Cognitive or Daily Variation. Neuron 98, 439-452.e5 (2018).
11. Finn, E. S. et al. Functional connectome fingerprinting: identifying individuals using
patterns of brain connectivity. Nat. Neurosci. 18, 1664–1671 (2015).
12. Shen, X. et al. Using connectome-based predictive modeling to predict individual
behavior from brain connectivity. Nat. Protoc. 12, 506–518 (2017).
(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 18, 2020. . https://doi.org/10.1101/2020.03.16.993709doi: bioRxiv preprint